An Extended TopoART Network for the Stable On-line Learning of Regression Functions

Tscherepanow, Marko

doi:10.1007/978-3-642-24958-7_65

Cited by 4 publications

(4 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…TopoART has spawned several variants for unsupervised [36][37][38], supervised [73,74], and semi-supervised [90] learning paradigms.…”

Section: Fuzzy Topoartmentioning

confidence: 99%

“…Distributed learning is also featured in the ART variants intro-duced in [66,67]. In the ART literature, the power of distributed activation has been harnessed to perform, for instance, (a) unsupervised feature extraction [68]; (b) hierarchical clustering [21,29] -although featuring distributed representation, the latter approaches are cascade architectures not designed to model arbitrarily-shaped clusters since they are limited by their category representations at each hierarchical level; and (c) supervised learning systems such as the distributed ARTMAP [65], which is a generalization of a variety of ART models [69] such as [15,[69][70][71][72] and uses distributed ART as its building block, some topoART variants [73,74], default ARTMAPs [69,71], and adaptive resonance associative map [9] variants [31,75].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Distributed dual vigilance fuzzy adaptive resonance theory learns online, retrieves arbitrarily-shaped clusters, and mitigates order dependence

2020

View full text Add to dashboard Cite

This paper presents a novel adaptive resonance theory (ART)-based modular architecture for unsupervised learning, namely the distributed dual vigilance fuzzy ART (DDVFA). DDVFA consists of a global ART system whose nodes are local fuzzy ART modules. It is equipped with the distinctive features of distributed higher-order activation and match functions, using dual vigilance parameters responsible for cluster similarity and data quantization. Together, these allow DDVFA to perform unsupervised modularization, create multi-prototype clustering representations, retrieve arbitrarily-shaped clusters, and control its compactness. Another important contribution is the reduction of order-dependence, an issue that affects any agglomerative clustering method. This paper demonstrates two approaches for mitigating order-dependence: preprocessing using visual assessment of cluster tendency (VAT) or postprocessing using a novel Merge ART module. The former is suitable for batch processing, whereas the latter can be used in online learning. Experimental results in the online learning mode carried out on 30 benchmark data sets show that DDVFA cascaded with Merge ART statistically outperformed the best other ART-based systems when samples were randomly presented. Conversely, they were found to be statistically equivalent in the offline mode when samples were pre-processed using VAT. Remarkably, performance comparisons to non-ART-based clustering algorithms show that DDVFA (which learns incrementally) was also statistically equivalent to the non-incremental (offline) methods of DBSCAN, single linkage hierarchical agglomerative clustering (HAC), and offline version of k-means, while retaining the appealing properties of ART. Links to the source code and data are provided. Considering the algorithm's simplicity, online learning capability, and performance, it is an ideal choice for many agglomerative clustering applications.Representation, Topology, Visual Assessment of Cluster Tendency. methods) and top-down (divisive or splitting methods) [2]. Hierarchical ART architectures generally follow two main designs [20]: (a) a series/cascade of ART modules where the output of one ART (i.e., a prototype) is the input of the next [21][22][23][24][25][26][27][28][29][30][31] or (b) parallel ART modules sharing the same inputs and using different vigilance values [6,[32][33][34][35][36][37][38][39]. Generally, the hierarchical relationships between ART modules are defined implicitly by the input signal flow, explicitly by enforcing constraints or connections, and/or by the setting of multiple vigilance parameters to define hierarchies. Alternatively, hierarchies within the same ART can be created by designing custom ART activation functions [40,41] or by analyzing its distributed activation patterns [42]. ART-based hierarchical approaches have been successfully applied, for instance, in text mining [20,43] and robotics [30,39].Another branch of clustering includes multi-prototype-based methods. These allow multiple prototypes to represent a si...

show abstract

“…TopoART has spawned several variants for unsupervised [36][37][38], supervised [73,74], and semi-supervised [90] learning paradigms.…”

Section: Fuzzy Topoartmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Distributed dual vigilance fuzzy adaptive resonance theory learns online, retrieves arbitrarily-shaped clusters, and mitigates order dependence

2020

View full text Add to dashboard Cite

show abstract

“…TopoART-R (Tscherepanow, 2011) is a variant of fuzzy topoART (Sec. 2.2.2) designed for regression purposes.…”

Section: Topoart-rmentioning

confidence: 99%

A survey of adaptive resonance theory neural network models for engineering applications

2019

View full text Add to dashboard Cite

This survey samples from the ever-growing family of adaptive resonance theory (ART) neural network models used to perform the three primary machine learning modalities, namely, unsupervised, supervised and reinforcement learning. It comprises a representative list from classic to modern ART models, thereby painting a general picture of the architectures developed by researchers over the past 30 years. The learning dynamics of these ART models are briefly described, and their distinctive characteristics such as code representation, long-term memory and corresponding geometric interpretation are discussed. Useful engineering properties of ART (speed, configurability, explainability, parallelization and hardware implementation) are examined along with current challenges. Finally, a compilation of online software libraries is provided. It is expected that this overview will be helpful to new and seasoned ART researchers. 7 Code repositories 54 8 Conclusions 55 2 Compute activation function(s): Tj = fT (x, θj, α), ∀j ∈ C. 3 Perform WTA competition: J = arg max j∈Λ (Tj). 4 Compute match function(s):7 Update category J: θ new J = fL(x, θ old J , β). 8 else 9 Deactivate category J: Λ ← Λ − {J}. 10 if Λ = {∅} then 11 Go to step 3. 12 else 13 Set J = |C| + 1. 14 Create new category: C ← C ∪ {J}. 15 Initialize new category: θ new J = fN (x, λ). 16 Set output: y where the vigilance criterion is M J ≥ ρ.Learning. If the winning category satisfies the vigilance test, then resonance occurs, and the radius R J and centroid m J of the winning node are updated as follows:where ρ ∈ (0, 1] is the vigilance parameter.Learning. If the winning category J satisfies M J ≥ ρ, then resonance occurs, and it is updated as follows:J J needs to be updated after the presentation of each sample, then γ k (new) can be estimated incrementally. Particularly, when there is a resonant committed node J, if x k is a meta-information feature, thenotherwise,

show abstract

“…For the prediction of the robots posture, a regression on top of a topology-learning ART (Adaptive Resonance Theory) neuronal network [9] is used. It uses the image vector directly to predict a 15-dimensional posture where each of the dimensions describes the position of one actuator in the range of [0..1].…”

Section: A Robotic Imitation Systemmentioning

confidence: 99%

Direct on-line imitation of human faces with hierarchical ART networks

Holthaus

Wachsmuth

2013

2013 Ieee Ro-Man

View full text Add to dashboard Cite

Abstract-This work-in-progress paper presents an on-line system for robotic heads capable of mimicking humans. The marker-less method solely depends on the interactant's face as an input and does not use a set of basic emotions and is thus capable of displaying a large variety of facial expressions. A preliminary evaluation assigns solid performance with potential for improvement. I. MIMICRY IN SOCIAL ROBOTICSMimicry of facial expressions (c.f.[1] for a comprehensive overview) is an automatic process that has been reported to occur at around five times a minute in human beings [2]. It is believed to facilitate face-to-face interaction between interlocutors and has positive effects for both the mimicker and the one being imitated [3]. For example, the affiliation between two interaction partners is being increased with the help of mimicry [4].To have robots engage in an interaction to a higher degree, and thus be socially more intelligent, it is desirable to give them the capability of mimicking human facial expressions. With such a system, a robot would be able to express a certain amount of empathy towards a human.A system enabling the robot to mimic a human has to work in an on-line fashion, and be robust plus flexible. Most existing mimicry-systems for robots today classify a fixed set of facial expressions (e.g. six basic emotions, [5]) and then generate the corresponding pre-modeled posture or a combination of them which is then displayed on the robot.[6] however, use a regression to learn direct motor mappings from facial images. This way, arbitrary postures can be displayed on the robot without the need to classify facial expressions beforehand. Also, the method does not need markers in the interactant's face or a feature point extraction because the raw image is used as an input for the regression. Therefore, one does not need to prepare an interaction partner in any way for the imitation to work. II. A ROBOTIC IMITATION SYSTEMIn [6], various methods for learning the relation between input image and posture have been shown to have comparable performances. Their results however are solely based on an off-line evaluation with pre-recorded set of images from the same camera. Our goal in this work has been to improve the proposed method and establish a system that has the same flexibility and accuracy as before but being implemented as an on-line system on a robot that runs in real-time and can be continuously improved by further learning.Our system is specifically trained for the robotic head "Flobi" [7] that has 15 degrees of freedom for facial expressions but can in principle be extended to other robots that have similar expressive capabilities. It relies on a standard face detection algorithm [8] which cuts out the face from the camera input image. The gray-scale image of the facial region is then histogram-normalized to floating point values between [0..1] and scaled to a 64x64 image. For the regression, parts with no information about the facial expression, mostly the lower corners of the image and no...

show abstract

An Extended TopoART Network for the Stable On-line Learning of Regression Functions

Cited by 4 publications

References 16 publications

Distributed dual vigilance fuzzy adaptive resonance theory learns online, retrieves arbitrarily-shaped clusters, and mitigates order dependence

Distributed dual vigilance fuzzy adaptive resonance theory learns online, retrieves arbitrarily-shaped clusters, and mitigates order dependence

A survey of adaptive resonance theory neural network models for engineering applications

Direct on-line imitation of human faces with hierarchical ART networks

Contact Info

Product

Resources

About